L’information quantique peut être traitée sur différents types de systèmes physiques. Elle peut également être traitée selon deux façons fondamentalement différentes, en usant soit des variables discrètes, soit des variables continues. Dans cette thèse, nous nous concentrons sur l’optique quantique en variables continues et les expériences étudiées sont basées sur l’utilisation de photons. Après une introduction à quelques notions de base de la mécanique quantique, nous présentons un protocole sous-universel d’informatique quantique, appelé Boson Sampling, précédé d’une partie exposant des éléments de théorie de la complexité nécessaires pour comprendre la preuve de supériorité quantique de ce protocole. Puis, nous proposons un modèle pour décrire la création de qudits intriqués dans une expérience réalisée au sein de notre équipe. Enfin, la dernière partie de cette thèse présente une interprétation physique plus fondamentale des résultats obtenus lors d’expériences de type Hong-Ou-Mandel avec des filtres en fréquence devant les photodétecteurs, et montre qu’il s’agit d’une partie d’un état chat de Schrödinger produit par post-sélection. / Quantum information can be processed on differents types of physical systems. It can also be processed in two fundamentally different ways, using either discrete or continuous variable implementations. In this thesis, we concentrate on quantum optics in continuous variables and the studied experiments are based on the use of photons. After an introduction to some basic notions of quantum mechanics, we present a subuniversal protocol of quantum computing, named Boson Sampling, preceded by a part exposing elements of complexity theory which are necessary to understand the quantum superiority proof of this protocol. Then, we propose a model to describe the creation of entangled qudits in an experiment done in the team. Finally, the last part of this thesis presents a more fundamental physical interpretation of the results obtained during Hong-Ou-Mandel experiments with frequency filters in front of the photodetectors, and shows that we have a part of a Schrödinger cat state, produced by postselection.

Photonics has consistently played an important role in the investigation of quantum-enhanced technologies and the corresponding study of fundamental quantum phenomena. The majority of these experiments have relied on the free space propagation of light between bulk optical components. This relatively simple and flexible approach often provides the fastest route to small proof-of-principle demonstrations. Unfortunately, such experiments occupy significant space, are not inherently phase stable, and can exhibit significant scattering loss which severely limits their use. Integrated photonics offers a scalable route to building larger quantum states of light by surmounting these barriers. In the first half of this thesis, we describe the operation of on-chip heralded sources of single photons. Loss plays a critical role in determining whether many quantum technologies have any hope of outperforming their classical analogues. Minimizing loss leads us to choose Spontaneous Four-Wave Mixing (SFWM) in a silica waveguide for our source design; silica exhibits extremely low scattering loss and emission can be efficiently coupled to the silica chips and fibers that are widely used in quantum optics experiments. We show there is a straightforward route to maximizing heralded photon purity by minimizing the spectral correlations between emitted photon pairs. Fabrication of identical sources on a large scale is demonstrated by a series of high-visibility interference experiments. This architecture offers a promising route to the construction of nonclassical states of higher photon number by operating many on-chip SFWM sources in parallel. In the second half, we detail one of the first proof-of-principle demonstrations of a new intermediate model of quantum computation called boson sampling. While likely less powerful than a universal quantum computer, boson sampling machines appear significantly easier to build and may allow the first convincing demonstration of a quantum-enhanced computation in the not-distant future. Boson sampling requires a large interferometric network which are challenging to build with bulk optics, we therefore perform our experiment on-chip. We model the effect of loss on our postselected experiment and implement a circuit characterization technique that accounts for this loss. Experimental imperfections, including higher-order emission from our photon pair sources and photon distinguishability, are modeled and found to explain the sampling error observed in our experiment.